httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From r..@ai.mit.edu (Robert S. Thau)
Subject Re: Suggested changes
Date Thu, 16 May 1996 19:28:40 GMT
Re: the memset in make_sub_pool --- that is zeroing out the pool
*header*, which is a fairly small struct, not the whole thing.  This
is reasonably fast, and also important (lest p->cleanups, say, start
off with some random value).

As for chk_on_blk_list, I would *not* want to see it removed entirely
--- we may need it again some day.  I would be a little friendlier to
the notion of #ifdeffing it out, but not all that much --- the last
time I ran a profile, that function accounted for 0.4% of the CPU time
taken up by the web server, and I think that the safety check it
provides is worth that much at least.  (Granted the errors it checks
for will only happen once in a blue moon, but they're of a sort which
is very difficult to detect unless you check for it explicitly, as I
found out the hard way when I wrote the stuff originally).

If speed bumming is a concern, a profile will probably discover much
better pickings elsewhere --- figuring out an intelligent way to cache
.htaccess information, or even diddling the default config files so that
people who used them straight up wouldn't have their servers looking for
/.htaccess on every request, would probably yield *much* greater benefits.

rs ("premature optimization is the root of all evil") t

Mime
View raw message